jinhongyii commented on code in PR #16094:
URL: https://github.com/apache/tvm/pull/16094#discussion_r1388405514
##########
tests/python/relax/distributed/test_distributed_transform_propagate_sharding.py:
##########
@@ -1060,16 +1738,19 @@ def test_mlp_pipeline_parallelism():
def test_decoder_layer():
- # mod = relax.transform.LegalizeOps({"relax.reshape": lambda bb, call:
bb.normalize(call)})(LlamaAttentionLayer)
- mod = LlamaAttentionLayer
- after = relax.distributed.transform.PropagateSharding()(mod)
+ after =
relax.distributed.transform.PropagateSharding()(LlamaAttentionLayer)
assert_structural_equal(after, ShardedLlamaAttentionLayer)
-def test_decoder_layer_dynamic_shape():
- # mod = relax.transform.LegalizeOps({"relax.reshape": lambda bb, call:
bb.normalize(call)})(LlamaAttentionLayer)
- mod = LlamaAttentionLayerDynamicShape
+def test_decoder_layer_tir():
+ mod = relax.transform.LegalizeOps()(LlamaAttentionLayer)
Review Comment:
This test checks whether this pass can propagate sharding annotations
through TIR, and TIR are not treated as opaque funcs. The shapes are not
changed because we are changing Tensor to DTensor which contains an extra field
`placement`. The reason that I use such a long test is to test whether the pass
can cover all of the common TIRs that appear in LLM.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
